A method of training a machine learning system. The method comprises collecting a first simulation dataset derived from a computer simulating a hypothetical scenario with a first simulation configuration having a first degree of fidelity. The method further comprises collecting a second simulation dataset derived from a computer simulating the hypothetical scenario with a second simulation configuration having a second degree of fidelity different than the first degree of fidelity. The method further comprises building a multi-fidelity training dataset including training data from both the first simulation dataset and the second simulation dataset according to an interleaving protocol.
Legal claims defining the scope of protection, as filed with the USPTO.
2. The method of claim 1, wherein the first and second degrees of fidelity are based on a user-defined importance score.
3. The method of claim 1, wherein collecting the first simulation dataset includes running the first simulation configuration at the first degree of fidelity.
4. The method of claim 1, wherein the first simulation configuration includes a variational autoencoder simulation.
5. The method of claim 4, wherein the variational autoencoder simulation is trained to emulate a different simulation configuration.
6. The method of claim 5, wherein training the variational autoencoder simulation includes supervised learning on a plurality of execution traces from another simulator configuration.
7. The method of claim 1, wherein the first simulation dataset and the second simulation dataset are two of a larger plurality of simulation datasets, wherein the multi-fidelity training dataset is built from training data from each of the plurality of simulation datasets.
8. The method of claim 1, wherein the multi-fidelity training dataset is built according to an interleaving protocol that sets the first inclusion frequency and the second inclusion frequency based at least on the first degree of fidelity and the second degree of fidelity.
9. The method of claim 8, wherein the interleaving protocol is a multi-armed bandit protocol.
10. The method of claim 8, wherein the interleaving protocol sets the first inclusion frequency and the second inclusion frequency further based on a cost-benefit trade-off between the first simulation configuration and the second simulation configuration.
11. The method of claim 8, wherein the interleaving protocol sets the first inclusion frequency and the second inclusion frequency further based on probabilistically sampling from one or both of the first simulation dataset in proportion to the first degree of fidelity and the second simulation dataset in proportion to the second degree of fidelity.
12. The method of claim 1, wherein the multi-fidelity training dataset is pre-cached for training the machine learning system.
13. The method of claim 1, wherein at least a portion of one or both of the first simulation dataset and the second simulation dataset included in the multi-fidelity training data is generated during training of the machine learning system.
14. The method of claim 1, wherein the machine learning system is configured for reinforcement learning.
15. The method of claim 1, wherein the machine learning system is configured for imitation learning.
16. The method of claim 1, wherein the machine learning system is configured for supervised learning.
17. The method of claim 1, wherein the machine learning system is configured for unsupervised learning.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
May 15, 2020
January 24, 2023
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.